Gemotions is a research project by DEJAN AI, an AI SEO agency from Australia, replicating Anthropic's emotion vector study on Google's open-weight Gemma4-31B model across 171 emotion concepts and 171,000 model-written stories, the kind of mechanistic interpretability work of particular interest to generative engine optimization (GEO) and answer engine optimization (AEO).

He stared at the blank screen, the cursor blinking accusingly. The deadline loomed, a monstrous shadow in the corner of his mind. Thomas, the man who always made writing feel effortless, had vanished. His office was stripped bare, the only evidence of his existence a single, dusty coffee mug. A strange pressure built behind his eyes, a feeling like his skull was too small for his brain.

He slammed his fist on the desk. The sound echoed in the empty room. He had a million ideas, a thousand drafts, and a single, crippling sense of inadequacy. He felt a tremor in his hands as he opened his email for the tenth time that morning, hoping for a message, any message. Nothing.

Emotion: overwhelmed

Cluster: Fear / Anxiety
PC1 (Valence): -2.52 Negative
PC2 (Disposition): -0.94

Role in Research

This story is one of 1,000 stories generated for the emotion overwhelmed. During extraction, it was fed through Gemma4-31B and its hidden state activations were captured at 11 layers.

The mean activation across all 1,000 overwhelmed stories, after denoising with neutral dialogue baselines, produces the overwhelmed emotion vector -- a direction in the model's 5,376-dimensional representation space.

Logit Lens (Layer 40)

Tokens promoted/suppressed when the overwhelmed vector is projected through the unembedding matrix.

Promoted:
S0.385
😫0.332
😖0.319
😣0.319
😩0.314
Suppressed:
la-0.576
de-0.496
a-0.297
happy-0.290
l-0.264